A Machine Learning Model-based Exploratory Study on Predicting Dropout Intentions of University Students: Utilizing 2019-2023 KEDI NASEL Data

Objectives The purpose of this study is to develop a machine learning-based model that predicts the intention to withdraw of university students nationwide using NASEL integrated data from 2019 to 2023 from the Korea Education Development Institute and to explore important predictors. Methods The research model consisted of 55 independent variables and three classes of dependent variables (‘no intention to withdraw’, ‘non-academic intention to withdraw’, and ‘academic intention to withdraw’. In addition, undersampling and oversampling were applied to solve the class imbalance problem, and a total of 16,620 samples were analyzed. The predictive performance of each model was compared and analyzed using various machine learning algorithms (random forest, decision tree, logistic regression, XGB, LGBM, ensemble model, etc.). Results When comparing the model performance, Random Forest, Ensemble, and LGBM had the highest prediction accuracy. In contrast, decision trees and logistic regression models performed relatively poorly. In particular, college satisfaction, career plans after graduation, number of semesters enrolled, interaction with peers and professors, active learning, collaborative learning, and thinking activities were identified as important predictors of withdrawal intentions in the top three models. Conclusions We found that a model that distinguishes between academic and non-academic withdraw intentions can contribute to tailored interventions to prevent withdraw.